Artificial Intelligence · Cross-listed from cs.LG
When Do Supervised UQ Ensembles Improve LLM Hallucination Detection? A Robustness Study
Mohit Singh Chauhan, Vipin Gyanchandani and Dylan Bouchard
Uncertainty quantification (UQ) methods are widely used for
hallucination detection in large
language models (LLMs) in closed-book settings where ground-truth evidence is unavailable at inference time. Prior work has proposed combining UQ signals via learned ensembles, but empirical investigations into the robustness of these ensembles are limited. We study a supervised ensembling framework that trains a classifier over heterogeneous UQ-based scorer outputs on a small, domain-specific
dataset of labeled LLM responses, then applies it to out-of-sample
hallucination classification without
retrieval, tools, or reference documents. Across four LLMs, nine
datasets, and three generation regimes (short-form QA, long-form generation, and code generation), we provide a systematic robustness analysis along three axes: sample efficiency, in-domain
dataset transfer, and generation regime dependence. We find that supervised ensembles outperform the best individual scorer in 30 of 32 settings, with gains realized from as few as 100 labeled instances. Ensembles retain most of their advantage in cases of in-domain transfer under distribution shift, outperforming the best non-ensemble scorer in 23 of 28 transfer settings. Sampling-based black-box ensembles are nearly as effective as full ensembles, while single-generation white-box ensembles offer limited benefit.
Read the full abstract ▾
Artificial Intelligence · Cross-listed from cs.LG
Beyond Static Interpretability: Anticipating Post-SFT Mechanisms from Pre-SFT Parameters for Better Tuning
Hang Chen, Jiaying Zhu and Wenya Wang
Mechanistic Localization bridges mechanistic interpretability and post-training optimization by isolating critical parameters via interpretative approaches and then guiding parameter-efficient Supervised Fine-Tuning (SFT) in a “locating-then-tuning” paradigm. However, due to the retrospective nature of mechanistic interpretability, directly interpreting pre-SFT models introduces misleading conclusions. Specifically for novel tasks, initially identified
neurons differ drastically from those governing the final model, introducing biases that actively disrupt SFT. To address this, we propose a forward-looking localization framework that accurately estimates the post-SFT interpretability state using only pre-SFT parameters and the target
dataset. Theoretically, we model SFT as a continuous parameter
evolution, leveraging Taylor expansion to rigorously bridge the post-tuning mechanistic objective with the pre-SFT model's dynamic gradients. Practically, we design dual-granularity (
neuron- and component-level) localization pipelines. Extensive experiments demonstrate that our approach not only provides superior SFT guidance but also exhibits robust performance and temporal scalability across increasing model sizes. This work transcends the fundamental limitation of traditional interpretability-its inability to identify task-critical mechanisms before they are trained-pioneering a predictive frontier that unites mechanistic interpretability with targeted optimization.
Read the full abstract ▾
Artificial Intelligence
Implicit Q-learning-bootstrapped ant colony optimization for maritime moving-target observation scheduling with agile satellites
He Wang, Junyu Wu, Yeye Liu et al. (+5 more)
Maritime moving-target observation scheduling with agile Earth observation satellites is a dynamic, sequence-dependent combinatorial optimization problem. Sea-surface targets move continuously, causing feasible observation windows to vary with target motion and satellite orbital geometry. The scheduler must jointly determine task selection, satellite assignment, observation-window selection, and observation ordering under time-window, attitude-maneuvering, and onboard-resource constraints. This paper proposes an implicit Q-learning-bootstrapped ant colony optimization method, termed IQACO, for multi-satellite maritime moving-target observation scheduling. Rather than directly learning a task-selection policy, IQACO embeds an offline implicit Q-learning module into constructive ant colony optimization to adaptively adjust the pheromone factor, heuristic factor, and evaporation rate. A compact search-state representation captures pheromone distribution, current and historical-best solution quality, and iteration progress. During online scheduling, ant colony optimization constructs feasible observation sequences, while the learned policy adjusts the search behavior according to the current search state. Experiments on 14 scenarios with different scales and satellite configurations show that IQACO consistently outperforms the compared
algorithms, improving the mean objective value over conventional ant colony optimization by 2.86%-9.41%. Further comparative and supplementary experiments demonstrate its effectiveness and robustness across different scheduling conditions and problem settings. These results indicate that offline value learning provides an effective adaptive search-control mechanism for constrained maritime moving-target observation scheduling.
Read the full abstract ▾
All authors He Wang, Junyu Wu, Yeye Liu, Yifan Zhou, Jie Zhang, Hui Li, Yanjie Song, Liang Li
Artificial Intelligence
Reinforcement Learning-Guided Evolutionary Policy Optimization for Preference-Adjustable Heterogeneous Agile Earth Observation Satellite Scheduling
He Wang, Junyu Wu, Hui Li et al. (+3 more)
Heterogeneous agile Earth observation satellite (AEOS) scheduling requires task selection, satellite assignment, and observation sequencing under satellite-dependent visibility windows, attitude maneuvering requirements, energy consumption, and onboard storage constraints. Since satellites differ in orbital access, maneuvering capability, and payload resources, the same task may have different feasible windows, transition costs, and resource-consumption patterns on different platforms, which increases the difficulty of unified modeling and efficient optimization. To address this problem, this paper proposes an evolutionary policy optimization framework for heterogeneous AEOS scheduling with preference-adjustable weighted objectives. In the modeling layer, assignment-based indirect encoding is combined with decoder-based equivalent-cost evaluation to retain satellite-dependent constraints while integrating task gain, energy saving, and load balance into an interpretable scalar utility. In the optimization layer, schedule decoding, population-based search, and online actor-critic operator control are decoupled, so that reinforcement learning selects high-level search operators rather than constructing schedules directly. Based on this framework, a reinforcement-learning-assisted operator-selection memetic evolutionary
algorithm (RLOSMEA) is developed to coordinate global exploration, feasibility recovery, and local refinement under a limited function-evaluation budget. Experiments on different heterogeneous AEOS scenarios show that RLOSMEA achieves higher overall weighted utility and more stable convergence than representative metaheuristic baselines. Sensitivity and learning-behavior analyses further confirm the robustness of the proposed method and the effectiveness of reinforcement-learning-guided operator selection.
Read the full abstract ▾
All authors He Wang, Junyu Wu, Hui Li, Yanjie Song, Witold Pedrycz, Liang Li
Artificial Intelligence
HMGCLIP: Heterogeneous Multi-Granularity Contrastive Learning for E-commerce Representation Learning
Qiuyu Zhu, Yi Gao, Zhichao Wan et al. (+1 more)
Although recent
Multimodal Large
Language Models (MLLMs) have advanced general product understanding, they implicitly encode product information into global
embeddings, thereby limiting their ability to capture fine-grained attributes. This limitation hinders performance in tasks requiring precise attribute discrimination, such as distinguishing subtle material differences among visually similar products. To address this challenge, we propose HMGCLIP, a unified
multimodal embedding framework. By constructing a heterogeneous hypergraph, we leverage hypergraph
topology to mine structure-aware hard negatives and align multi-granular semantics at both relation and hyperedge levels. This design enables a dual-granularity inference mechanism that dynamically fuses attribute evidence for both fine-grained and coarse-grained downstream tasks. Furthermore, we release a comprehensive fine-grained e-commerce
dataset to facilitate future benchmarking. Extensive experiments on this new
dataset and the public MAVE
benchmark show that HMGCLIP outperforms strong
multimodal encoders, MLLMs, and e-commerce baselines, validating the superiority of HMGCLIP.
Read the full abstract ▾
All authors Qiuyu Zhu, Yi Gao, Zhichao Wan, Mingyang Ma
Artificial Intelligence
Mahalanobis-Based Multi-Head Attention for Complex State Propagation
Xiaohe Li
In this paper, we propose Mahalanobis-Based Multi-Head
Attention (MHA-CSP), a novel
attention mechanism that replaces the standard dot-product with a Mahalanobis distance-based RBF kernel, which effectively computes
attention in an infinite-dimensional feature space without increasing the parameter count. Crucially, the positive definiteness of the Mahalanobis distance enables a direct construction of Tree
Attention:
attention scores are built directly from accumulated distances, with a LogSumExp correction that rectifies the raw distance by subtracting the log-sum of edge exponentials. Moreover, the multi-head Mahalanobis distance
matrices are themselves repurposed to construct an
attention meshing mechanism, enabling cross-head kernel collaboration that simultaneously boosts accuracy and training efficiency. Extensive experiments demonstrate that MHA-CSP, with only 119K parameters and teacher forcing applied exclusively at the final hidden state, consistently outperforms
Transformer and GCN baselines trained from scratch under identical conditions on long-sequence state tracking tasks. While these baselines rely on dense
attention or graph propagation, MHA-CSP achieves robust structured
reasoning via synthetic distance rectification—powered by Mahalanobis-based
attention—and efficient information bypass inherited from the CSP backbone. This result highlights the effectiveness of complex-valued state propagation with collaborative multi-head rectification in capturing symbolic structures, establishing a new efficiency-performance trade-off for structured
reasoning.
Read the full abstract ▾
Artificial Intelligence
A Behavior-Guided Online Probabilistic Forecasting Method for Electric vehicle Charging Loads
Chenghan Li, Qingxiang Liu, Yinliang Xu et al. (+1 more)
Electric vehicle (EV) charging loads exhibit strong behavioral heterogeneity and temporal variability, posing significant challenges for online probabilistic
forecasting under evolving operating conditions. In particular, persistent charging patterns may differ substantially across stations, while recent behavioral changes can continuously alter the underlying load distributions. This paper proposes a behavior-guided online probabilistic
forecasting framework that explicitly characterizes persistent station-specific patterns and recent behavioral changes. A dual-timescale behavior representation is constructed to distinguish long-term charging characteristics from recent behavioral states and quantify their deviations. These behavioral changes are further semantically encoded to guide drift-aware
forecasting adaptation, while a delayed-feedback mechanism ensures temporally consistent online updates when observations become available across different
forecasting horizons. Experiments on ten heterogeneous real-world charging stations demonstrate that the proposed method consistently outperforms conventional
forecasting models and concept-drift-aware online baselines in
forecasting accuracy and probabilistic reliability. For 1-h-ahead
forecasting, the proposed method reduces MSE and Pinball loss by 15.3% and 17.8%, respectively, over the corresponding best baselines. For 4-h-ahead
forecasting, the improvements further reach 16.8% and 22.6%, respectively, demonstrating consistent performance gains under evolving charging behaviors and extended
forecasting horizons.
Read the full abstract ▾
All authors Chenghan Li, Qingxiang Liu, Yinliang Xu, Yuxuan Liang
Artificial Intelligence · Cross-listed from cs.LG
Evaluating Deep Multivariate Imputation Models on Wearable Device Data
Skye Goodman, Roussel Desmond Nzoyem, Leandro Junges et al. (+5 more)
Wearable device data enables continuous health monitoring, but suffers from structured missingness: features sharing a physical sensor drop out together. Deep imputation methods such as BRITS and SAITS have seen limited evaluation on
multimodal physiological data under realistic missingness, and existing
benchmarks use random-point holdout protocols that incorrectly assume missingness is independent across features and time. Using data from a person with epilepsy recorded on a Garmin smartwatch, we develop an evaluation protocol that mines contiguous missing-run templates from training data, stratifies them by per-feature gap-length
quantiles, and injects them as block masks with preserved co-missingness structure. A matched training protocol exposing models to the same missingness distribution reduces BRITS's severe-gap MAE by 43%, demonstrating the potential benefit of the proposed evaluation and training protocol within this single-participant
dataset. We further extend BRITS with time-of-day encoding and a circadian harmonic channel. No single model dominates: linear interpolation is optimal for slow-moving features over short gaps; extended BRITS achieves lower MAE on dynamic cardiac features in moderate and severe gaps; and SAITS better preserves the ground-truth distribution by Jensen-Shannon distance despite higher MAE. Ultimately, model rankings strongly depend on evaluation designs. By exposing how traditional evaluation methods obscure true model capabilities, our transferable protocol establishes critical steps towards developing better imputation strategies for future multi-sensor wearable
datasets.
Read the full abstract ▾
All authors Skye Goodman, Roussel Desmond Nzoyem, Leandro Junges, Peter Kissack, Yasser Qureshi, Amberly Brigden, Jeff Clark, Nawid Keshtmand
Artificial Intelligence
Partial Identification under Causal Orders by Linear Programming
Eric Rossetto and Alessandro Antonucci
Non-parametric (partial) identification of counterfactual queries typically relies on a fully specified causal graph. Motivated by settings with incomplete domain knowledge, we challenge this requirement by leveraging structural assumptions that are inherently implied by the query itself. We show that any counterfactual inquiry induces a, mostly partial, topological ordering over relevant variables, which, in turn, enables an explicit query parametrisation reducing the identification task to a linear program. This allows bounding arbitrary counterfactual and nested counterfactual queries. Our work can be viewed as a generalisation of the classical bounding framework of Tian and Pearl (2000), originally developed for probabilities of causation. We also prove the tightness of our bounds by constructing structural causal models that attain the bounds whilst being compatible with both the observed data and the query-implied order. To assess both the generality and practical utility of the proposed bounding procedure, we revisit several case studies from the literature, demonstrating how the derived bounds can be used to yield informative insights even in the absence of an input causal graph.
Read the full abstract ▾
Artificial Intelligence
A Judge Should Know What Changed:Construct Validity for LLM-as-a-Judge Evaluation
Jianlin Chen, Wenhui Chen, Ziyao Lin et al. (+1 more)
LLM-as-a-judge evaluation is usually assessed by agreement and robustness to surface perturbations, but reliability does not establish construct validity. We formalize construct validity for an evaluator as a two-dimensional profile: invariance S, the probability that a verdict is unchanged under construct-preserving edits, and construct sensitivity R, the probability that it changes under minimal construct-changing edits. We show that S and R are independent and that no scalar summary preserves all relevant comparisons. We measure the profile across 7 judges and 4 domains using 7 construct-changing intervention types and 5 register-only controls, with intervention direction determined by human annotators and generation, verification, and judging assigned to disjoint model families. At matched invariance S >= 0.90, judges average S = 0.945 but R = 0.319. Sensitivity also differs between scope and strength edits: R_scope = 0.383 versus R_strength = 0.262, a +0.121 gap with the same sign for all 7 judges. We further audit five public label sets and find that surface-only predictors reproduce 55%-67% of labels in paired mode, including 67.4% of MT-Bench human votes. These results show that high judge agreement can coexist with weak sensitivity to changes in the construct being evaluated, motivating joint reporting of invariance and sensitivity and
auditing the validation set itself.
Read the full abstract ▾
All authors Jianlin Chen, Wenhui Chen, Ziyao Lin, Chi Man Vong
Artificial Intelligence
ResiSpec: Enhancing Multi-Candidate Speculative Sampling via Residual Distribution Shaping
Zhi-Kai Chen, Jun-Jie Tao, Wei-Xiang Mao et al. (+2 more)
The efficiency of Large
Language Model (LLM) serving is fundamentally limited by the sequential nature of autoregressive decoding. Speculative Decoding (SD) mitigates this by using a lightweight draft model to speculate future
tokens, which are then validated by the LLM in a single parallel forward pass. To further boost efficiency, multi-candidate schemes propose diverse candidate sets to increase the likelihood of
token acceptance. However, we show that these schemes are bottlenecked by Residual Drift: a phenomenon where the rejection of initial candidates causes the residual target distribution to diverge from the draft model's predictions. This shift renders subsequent candidates ineffective and forces the system into expensive resampling. To resolve this, we propose ResiSpec, a framework that strategically reforms the proposal distribution during verification to anchor the residual target mass within the draft model's high-confidence regions. By mathematically re-aligning the verification process without compromising output exactness, ResiSpec prevents candidate obsolescence and achieves up to 1.92$\times$ speedup over state-of-the-art multi-candidate methods. Code is available at https://github.com/Czzzk/Resispec.
Read the full abstract ▾
All authors Zhi-Kai Chen, Jun-Jie Tao, Wei-Xiang Mao, De-Chuan Zhan, Han-Jia Ye
Artificial Intelligence · Cross-listed from cs.GT
Multilevel Fair Allocation under Additive Preferences
Maxime Lucet, Nawal Benabbou, Aurélie Beynier et al. (+1 more)
We study multilevel fair resource allocation with tree-structured hierarchical relations among
agents. At each level, the problem can be viewed locally as allocating an
agent's bundle to its children, the overall allocation being a trace of this process iterated down to the leaves. Assuming that internal nodes' utilities are the utilitarian welfare of their children, and the leaves have classical additive utilities over items, we first propose multilevel adaptations of usual envy-based
fairness notions (e.g., WEF1). We present three adaptations and show that the choice among them is not neutral. We prove that, under identical preferences, the three adapted envy-based notions coincide, and that the Multilevel extension of Weighted Round Robin (Chakraborty et al., 2021) (MWRR) guarantees them. We then show that under general preferences, MWRR may guarantee some notions while failing others. Finally, through experiments, we show that MWRR may still perform well even for adaptations it does not formally guarantee.
Read the full abstract ▾
All authors Maxime Lucet, Nawal Benabbou, Aurélie Beynier, Nicolas Maudet
Artificial Intelligence · Cross-listed from cs.LG
Equivariant Covariance Tensors: Guaranteed SPD Uncertainty for Tensor-Valued Geometric Learning
Ruihan Liu, Yu Ji, Jianbo Yu et al. (+2 more)
Tensor-valued prediction is fundamental to
geometric deep learning, yet uncertainty quantification (UQ) for such outputs remains an open challenge. While E(3)-equivariant neural networks excel at point estimates, they lack rigorous confidence measures. We focus on symmetric rank-2 tensor prediction, where the target has six Kelvin–Mandel coordinates and full uncertainty is represented by a $6\times6$ covariance
matrix. We introduce a framework for E(3)-equivariant UQ, modeling the full predictive distribution where both mean and covariance preserve rotational
symmetry. Our approach decomposes the covariance into irreducible representations $\mathrm{Sym}^2(\rho_c) \cong 2\times(l=0) \oplus 2\times(l=2) \oplus 1\times(l=4)$. By mapping from the flat Lie
algebra $\mathfrak{sym}(6)$ to the curved SPD
manifold via
matrix exponentiation, we strictly ensure positive-definite covariances while maintaining exact equivariance. Furthermore, we formulate a Log-Euclidean Equivariant Scoring Objective (LE-ESO)—a robust surrogate loss based on the Multivariate Laplace distribution—providing robustness to heavy-tailed errors and stable optimization. Validation on ModelNet40 inertia tensors and Materials Project dielectric tensors demonstrates that our method achieves competitive performance and provides physically consistent,
symmetry-preserving uncertainty estimates with useful risk and OOD sensitivity.
Read the full abstract ▾
All authors Ruihan Liu, Yu Ji, Jianbo Yu, Shifu Yan, Qingchao Jiang
Artificial Intelligence · Cross-listed from cs.CV
Markerless Pose Estimation for Resistance Training Technique Assessment
Joseph Turner, Jeff Clark and Nawid Keshtmand
Resistance training can be a high risk activity, and safe form is essential to avoiding injury. Laboratory-based movement analysis provides quantitive technique assessment, yet is not easily accessible. Markerless pose estimation infers body landmarks from images or
video without physical markers and could offer a feasible alternative for technique assessment. We present a pose estimation framework to evaluate resistance-training technique from ordinary
video footage. Using BlazePose, anatomical landmarks were extracted from squat, bench press, and deadlift
videos and converted into joint-angle trajectories, with the squat serving as the primary case study. Trajectories were assessed against a defined reference repetition using root mean square error (RMSE). Results show that the framework recovers meaningful kinematic patterns for the squat and deadlift, enabling quantitative comparison between repetitions and identification of technique variability within a set. Performance depended strongly on camera orientation and visual occlusion, with non-sagittal views distorting 2D joint-angle estimates. The findings demonstrate that markerless pose estimation can support accessible biomechanical assessment outside laboratory environments.
Read the full abstract ▾
Artificial Intelligence
Do Recipes Have Personas? Characterizing and Generating Creator Style in Attributed Procedural Graphs
Lei Jiang
While large
language models (LLMs) possess vast zero-shot procedural knowledge, their tendency to produce homogenized logic often obscures the unique, idiosyncratic execution processes of individual human creators. In this paper, we investigate the computational discovery of procedural personas from unstructured data. To achieve this, we introduce ViralRecipesTrans, a new
dataset of procedurally aligned execution flow graphs extracted from popular culinary
video transcripts and explicitly mapped to specific creators. We formulate procedural stylometry as a graph learning and process discovery task, revealing a fundamental duality: while traditional lexical classifiers overfit via semantic leakage, discrete topological metrics successfully capture the rigid physical constraints of a creator's workflow. Building upon this characterization, we extend our framework into a novel generative task–predicting a creator's exact structural execution graph for unseen dishes. We expose a fundamental dichotomy in style generation between global macro-planning and local structural execution. Our results demonstrate that few-shot LLMs dominate semantic assignment but suffer from persistent macro-planning deficits, whereas our structured two-stage model achieves superior topological control via rigid Markovian priors. Together, an ensemble approach to procedural generation combines the strengths from both sides, dynamically synthesizing global semantic
reasoning with localized topological footprints to automate the discovery and generation of personalized workflows.
Read the full abstract ▾
Artificial Intelligence
From State to Action: OODA-Tool for Reliable Multi-Turn Tool Use
Rongfeng Guo, Yinxuan Huang, Yusen Wu et al. (+5 more)
Reliable multi-turn tool use requires an
agent to preserve an evolving task state and ensure that each action remains consistent with it. However, direct function-calling and ReAct-style policies learn state tracking and action generation within the same autoregressive trajectory. This coupling creates state-action competition: the pressure to produce the next call can overwrite or ignore information accumulated earlier in the interaction. Inspired by Boyd's Observe-Orient-Decide-Act cycle, we introduce OODA-Tool, a typed closed-loop policy designed to mitigate this competition by separating state preservation from action realization. Rather than generating an action directly from the interaction history, OODA-Tool routes each decision through controller-checked intermediate states, ensuring that the final output remains grounded in the current task state. Specifically, Observe reconstructs the task state, Orient determines whether execution is warranted, Decide forms an admissible action structure, and Act realizes the external output. We evaluate OODA-Tool against direct function-calling and ReAct policies using Qwen3 models ranging from 0.6B to 14B across multi-turn, multi-tool, and incomplete-information settings. OODA-Tool consistently improves task success across model sizes, with larger gains on smaller models and on tasks whose actions depend strongly on information accumulated across turns and prior tool results. Controlled variants, stage-level ablations, and transfer evaluations further demonstrate the robustness of these improvements.
Read the full abstract ▾
All authors Rongfeng Guo, Yinxuan Huang, Yusen Wu, Maoqing Zhong, Yunlu Chen, Meng Tang, Teng Long, Vincent Tao Hu
Artificial Intelligence
Adaptive Influence Graphs for Failure Attribution in Multi-Agent Systems
Yarden Bakish, Amir Dudai, Roy Ganz et al. (+4 more)
Multi-
agent LLM systems are increasingly deployed in real-world applications, where failures can be costly and difficult to localize. Despite growing efforts to automate failure attribution, diagnosing failed runs still largely relies on human engineers. Yet engineers rarely debug complex systems by reading raw logs end to end. Instead, observability tools organize traces around components, actions, and dependencies to support targeted
navigation. We hypothesize that modern LLMs can benefit from the same paradigm. To test this hypothesis, we introduce Adaptive Influence Graphs (AIGs), a two-stage agentic framework that first transforms a failed trace into a structured graph and then navigates it to identify the critical error. Across multiple models, we show that richer trace representations consistently improve failure attribution, with adaptive graph construction and
agent-directed traversal yielding the strongest results. AIGs establish a new state of the art on Who&When, the standard
benchmark for multi-
agent failure attribution. This affirms our hypothesis that attribution depends not only on the diagnosing model, but also on how the trace is represented and explored.
Read the full abstract ▾
All authors Yarden Bakish, Amir Dudai, Roy Ganz, Oren Nuriel, Elad Ben Avraham, Mor Shpigel Nacson, Ron Litman
Artificial Intelligence
The Handoff Tax: Continuing Non-Native Trajectories in LLM Agents
Roy Ganz, Mor Shpigel Nacson, Adi Kalyanpur et al. (+1 more)
Coding
agents perform long-running tasks spanning dozens of model calls, tool uses, and code edits. As these runs unfold, users face a practical cost-quality trade-off: escalating to a stronger model when a cheaper one struggles, or downshifting once the hard
reasoning is complete. Each switch requires the receiver to continue a non-native trajectory produced by another model. We study how this handoff affects quality and cost, and how varying the trajectory information inherited by the receiver changes the outcome. Using pairs of low-cost, low-capability (LC) and high-cost, high-capability (HC) models from the Claude and GPT families, we vary handoff direction, timing, and interface, comparing full-trajectory transfer, compaction, and trajectory removal while preserving the repository state. Across both model families, full-trajectory escalation recovers less than half of the LC-to-HC quality gap while incurring a substantial cost premium. We term this cost-quality penalty the handoff tax. By contrast, downshift offers a favorable cost-quality point. Interestingly, the preferred interface also reverses with direction: reducing LC-model trajectory information improves escalation quality, whereas removing the HC-model trajectory reduces downshift quality.
Read the full abstract ▾
All authors Roy Ganz, Mor Shpigel Nacson, Adi Kalyanpur, Ron Litman
Artificial Intelligence · Cross-listed from cs.CR
Not All Tokens Are Equal: Region-Aware Consistency Repair of Backdoors in MLLMs
Jiali Wei, Ming Fan, Mingkun Zhang et al. (+6 more)
MLLMs are increasingly deployed in user-facing applications, yet they inherit
backdoor risks from the pipelines used to construct them: triggers may reside in images, texts, or both. Existing model-level
backdoor removal methods, largely designed for conventional classifiers, show limited effectiveness on MLLMs, while MLLM-specific defenses mainly operate at inference time, filtering suspicious inputs without removing the
backdoor embedded in the model. To address this gap and eliminate latent
backdoors from MLLMs at their source, we present RACER, a model-level repair framework motivated by a key observation:
backdoors induce abnormal layer-to-layer
evolution in internal representations, which we term the layer-wise inconsistency anomaly. Importantly, this anomaly is modality-dependent, concentrating primarily in the
token region encoding the trigger features that the
backdoor model actually relies on. RACER therefore decomposes the fused representation into visual and textual
token regions, normalizes their layer-wise inconsistency separately, and recomposes them using modality-aware weights over a deep-layer window, yielding a region-aware inconsistency objective that better captures localized
backdoor-induced anomalies. Through a min-max optimization, this objective drives worst-case perturbation synthesis and
adversarial fine-tuning against the resulting perturbation to repair the model, suppressing the deep representational directional shifts on which
backdoor behaviors rely. RACER requires only 100 clean samples and no knowledge of the trigger, attack objective, or even whether the input model contains a
backdoor. Evaluations on three open-source MLLMs across 36
backdoor settings spanning image, text, and
multimodal triggers show that RACER reduces the average ASR to 1.1%, reaching 0% in 32 settings, while preserving clean-task utility on both
backdoor and clean models.
Read the full abstract ▾
All authors Jiali Wei, Ming Fan, Mingkun Zhang, Haoyu Wang, Jun Sun, Guoheng Sun, Xiaoning Ren, Haijun Wang, Ting Liu
Artificial Intelligence · Cross-listed from cs.CL
FARCA: Fact-Aligned Reliability-Aware Credit Assignment for Reinforcement Learning with Factual Supervision
Qiming Xie, Wenjie Zheng, Xiangqing Shen et al. (+1 more)
To reduce the
hallucination risk caused by outcome-driven rewards in large
language models trained through reinforcement learning with verifiable rewards, existing mitigation approaches introduce process-level factual supervision. However, due to coarse-grained aggregation of factual signals and the lack of reliability assessment for these signals, they create a mismatch between fact verification and policy updates. We term this noisy factual credit assignment and decompose it into two aspects: credit localization ambiguity and credit reliability ambiguity. To address these issues, we propose FARCA (Fact-Aligned Reliability-Aware Credit Assignment), a policy optimization framework that transforms factual supervision into localized, reliability-weighted
token-level training signals. FARCA achieves fine-grained credit localization by aligning the granularity of fact verification with that of policy updates. It further introduces counterfactual evidence attribution, which uses the dependence of a factual judgment on key evidence as an empirical proxy for verification reliability to compute reliability weights. These weights modulate factual rewards and local policy advantages, reducing the influence of potentially unreliable signals on policy optimization. Experiments across different models and multiple factual
reasoning benchmarks show that FARCA significantly improves model factuality while preserving general
reasoning capabilities.
Read the full abstract ▾
All authors Qiming Xie, Wenjie Zheng, Xiangqing Shen, Rui Xia
Artificial Intelligence · Cross-listed from cs.CV
Metadata-Aware Adaptation of a Generative Foundation Model for Conditional CMR Synthesis
Marc Rodríguez, Grzegorz Skorupko, Nay Aung et al. (+3 more)
Synthetic image generation is a promising strategy to address data scarcity and the underrepresentation of clinically important phenotypes in medical imaging, yet generating images that faithfully reflect meaningful patient characteristics remains challenging. In this work, we investigate metadata-conditioned cardiac
magnetic resonance (CMR) synthesis using a pretrained latent
diffusion model, encoding structured clinical metadata and slice position as textual
prompts to guide CMR generation. To improve metadata adherence and address the imbalance of clinical attributes, we integrate three strategies: Metadata-Free Classifier-Free Guidance (CFG), Contrastive Batching, and Inverse-Frequency Sampling. The framework was fine-tuned and evaluated on 59,058 short-axis CMR from the UK Biobank using paired image similarity, distributional fidelity, and subgroup-level analyses. The combined approach achieved a Fréchet Inception Distance (FID) of 37.47, improving by 57.04% over the same model fine-tuned without these strategies and by 28.68% over a previous text-conditioned CMR
diffusion baseline requiring cardiac geometry as additional input, while relying solely on patient metadata. This distributional gain, driven mainly by Metadata-Free CFG, came with a modest reduction in paired similarity, suggesting that the model prioritizes population-level realism over exact image reproduction. Subgroup analyses demonstrated improved
alignment across demographic and acquisition-related metadata, with disease-specific conditioning being the most challenging task. These findings demonstrate the potential of generative foundation models for clinically meaningful CMR synthesis while highlighting the need for more effective metadata-aware conditioning strategies. Our code is available at https://github.com/rodriguezmarc/conditional-cmr.
Read the full abstract ▾
All authors Marc Rodríguez, Grzegorz Skorupko, Nay Aung, Steffen E Petersen, Karim Lekadir, Polyxeni Gkontra
Artificial Intelligence · Cross-listed from cs.CV
Mind the Student: Behavioral and Contextual Cues for Automated Engagement Prediction in Online Learning
Alperen Kantarci, Visvanathan Ramesh and Gemma Roig
The prediction of student engagement from the online tutoring
videos is difficult because engagement is a multidimensional construct comprising distinct behavioral, emotional, and cognitive states. A reliable prediction requires bringing together different types of behavioral signals as well as expressive cues. Through our analysis of the CASED
dataset, it is clear that engagement prediction gets even harder due to the high inter-person variability as well as the subjectivity of the engagement annotation. To tackle these challenges, we develop a
multimodal framework that integrates the implicit spatiotemporal features extracted from pretrained
video, audio, and image encoders along with structured behavioral modalities like head pose, gaze, facial action units,
emotion, and wavelet-based audio features. We integrate these modalities via a Perceiver IO latent bottleneck. Moreover, student and instructor personalities are modeled as variational posteriors over learnable
embeddings to enable partial pooling across participants. We employ evidential regression and spectral-normalized Gaussian process classification heads for uncertainty-aware prediction to further improve robustness and calibration.
Benchmark on the CASED challenge test set shows that all participating methods converge near random-chance performance, revealing the difficulty of the
dataset. In this highly ambiguous regime, our framework achieves competitive performance while uniquely offering well-calibrated uncertainty metrics, demonstrating that reliable risk-quantification is an essential prerequisite for deploying engagement models in real-world educational tools.
Read the full abstract ▾
Artificial Intelligence
Selective Regenerative Decoding: Trajectory-Level Intervention for Inference-Time Reasoning
Sophia Xiao Pu, Yumo Xu, Sailik Sengupta et al. (+5 more)
Inference-time decoding methods improve LLM
reasoning by exploring multiple candidate trajectories, yet treat each trajectory as atomic: either retaining it whole or discarding it irreversibly. This wastes computation on partially promising candidates whose high-quality prefixes are abandoned alongside degraded suffixes. We introduce Selective Regenerative Decoding (SRD), which routes each candidate to discard, keep, or refine only the degraded portion of the suffix while preserving the useful prefix of borderline candidates, without requiring a larger target model. Under mild assumptions, SRD achieves a provable 1.28-to-1.36-fold gain in sample efficiency over rejection sampling with strictly higher expected trajectory quality, with the gain growing as the candidate pool grows. Across MATH500, GPQA Diamond, HotpotQA, and AlpacaEval with multiple generation-reward model pairs, SRD matches Best-of-N accuracy with substantially fewer generated
tokens and outperforms speculative rejection in low-compute regimes. By enabling segment-level intervention rather than whole-trajectory selection, SRD opens a previously underexplored region of the accuracy-compute tradeoff for inference-time
reasoning.
Read the full abstract ▾
All authors Sophia Xiao Pu, Yumo Xu, Sailik Sengupta, Millennium Bismay, Ruixue Lian, James Gung, Yi-an Lai, Arshit Gupta
Artificial Intelligence
SonarLLM: A Native Sonar–Optical Multimodal Large Language Model for Underwater Perception
Cong Su, longxuan ma, Ling Dong et al. (+4 more)
Reliable
underwater perception requires complementary sensing under variable visibility. Optical cameras capture appearance and semantics but degrade rapidly with turbidity, whereas imaging sonar preserves geometry while exhibiting distinct range-azimuth structure and acoustic artifacts. Existing MLLMs, built primarily on optical encoders, are therefore ill-suited to model sonar or adaptively exploit sonar-optical complementarity. We propose SonarLLM, a sonar-optical MLLM that treats sonar as a native perceptual modality. It combines a sonar-specific encoder, modality-specific physics-aware feature enhancement, and reliability-aware hierarchical fusion to align acoustic structure with optical semantics and dynamically adjust their contributions as sensing quality changes. We also introduce SonarBench, a paired
benchmark that spans four tasks: recognition, counting, visual question answering, and captioning; and, across the
benchmark, three input settings: sonar-only, optical-only, and fusion. By fixing the scene and sonar observation while varying optical degradation, SonarBench enables controlled measurement of cross-modal complementarity. SonarLLM achieves 72.0% macro accuracy across sonar-only recognition, counting, and VQA, outperforming the strongest baseline by 34.4 percentage points, and 68.7% under fusion, exceeding the best baseline by 25.1 points. For recognition and counting, the fusion-over-optical gain grows from 6.0 to 36.0 points as turbidity increases, indicating the increasing complementary value of sonar under controlled optical degradation. Together, these results show that robust heterogeneous perception depends not only on adding sonar, but on representing and weighting it according to its sensing characteristics.
Read the full abstract ▾
All authors Cong Su, longxuan ma, Ling Dong, Guofeng Tang, Weijie Yin, Haohui Chen, Zhengtao Yu
Artificial Intelligence
Can a Dynamic Internal Field Govern a Transformer's Cognition? Certifiability, not Superiority, in Homeostatic Compute Control
Francisco M. Arrabal-Campos, Ignacio Fernandez, Francisco G. Montoya et al. (+1 more)
An intelligent system does not merely reason: it governs its own
reasoning - how much to compute, when to stop, which module to activate. Can that role be played by a dynamic internal field - a low-dimensional homeostatic state with explicit physics and certified stability - that modulates cognition without performing it? Ours is a field on the module graph governed by a family of PDEs on the graph
Laplacian, advancing with an adaptive-depth reasoner. We certify the stability of the integrator of the whole family - an integrator certificate, not a closed-loop one. New, and proved here: a discrete Schur-Cohn criterion for Verlet with velocity coupling, necessary and sufficient per latent root, with no commutation hypothesis. The answer is threefold: substance no, structure only in part, certifiability yes. The type of the field's physics is irrelevant for accuracy: wave,
diffusion, gated mixtures and a 2D Navier-Stokes substrate tie. A twenty-seed preregistered deconfounding campaign bounds the structural claim: at equalized caps the second-order effect is strong in one family (+0.087 [+0.042, +0.132], t=4.0) but is not detected in the other (+0.014 [-0.013, +0.040], n.s.), so part of the original contrast was capacity, not order; and a matched-interface GRU is indistinguishable in the first and nominally exceeds the field in the second (-0.035 [-0.067, -0.002]). What distinguishes the field is not capability but that its one-step operator admits an exact runtime stability check - a difference of kind, not of existence: learned recurrences carry certificates too, sufficient and conservative ones. A kill-gate with a positive control finds no evidence for the field as evidence accumulator (Delta AUC +0.0007 [-0.0065, +0.0079] vs a 0.03 threshold). A dynamic internal field is a viable, certifiable compute governor, but not an enhancer of cognition: it modulates, it does not think.
Read the full abstract ▾
All authors Francisco M. Arrabal-Campos, Ignacio Fernandez, Francisco G. Montoya, Alfredo Alcayde
Artificial Intelligence
Benchmarking LLM Judges for Voice-Agent Evaluation: Reliability, Calibration, and Human Oversight
Anupam Purwar, Shashank Singh and Kritika Srivastava
Evaluating conversational voice
agents at scale re- quires reliable assessment methods that capture both observ- able interaction quality and the contextual judgment typically provided by human evaluators. We investigate LLM-as-a-Judge evaluation by comparing human judgments with GPT-4.1 and GPT-5 on telecom and retail voice-
agent conversations, across conversational quality and safety dimensions. The same interac- tions are scored under three evaluation configurations, p0, p1, and p2, to test whether automated judgments are sensitive to the evaluation setup and whether observed patterns generalize across configurations and judge models. Beyond aggregate agreement, we examine metric-level correlations, evaluator consistency, and systematic human-LLM disagreement to identify which conver- sational attributes can be judged reliably by automation and which remain sensitive to interpretation and context. Effective voice-
agent evaluation is also shaped by pipeline-level factors such as
speech generation, streaming, and error propagation across ASR,
reasoning, and tool-calling stages, motivating our focus on comparing how human and LLM judges score the same interactions end to end. Our results show that LLM- based evaluation can serve as an effective component of large- scale voice-
agent assessment, but that its reliability is metric- and configuration-dependent rather than uniform. This pro- vides an empirical framework for identifying which metrics suit automated evaluation and supports hybrid pipelines in which LLM judges handle scalable assessment while human evaluators remain engaged for metrics that demand contextual interpretation and higher-confidence judgment.
Read the full abstract ▾
Artificial Intelligence
OPDSearch+: On-Policy Distillation with RL Refinement for Search-Augmented Reasoning
Qinglin Ye, Zhiyuan Gu, Jingjie Xia et al. (+6 more)
Search-augmented
reasoning remains difficult for small
language models. On-policy
distillation (OPD) from trained teachers offers a promising direction, but suffers from two issues: (1) high-quality multi-turn search trajectories depend on dynamic retriever responses, making SFT data prohibitively expensive to collect at scale; (2) task-specifically trained teachers incur substantial training cost, while directly applying OPD with an off-the-shelf teacher without task-specific fine-tuning constrains the student to the teacher's performance ceiling and suffers from severe training instability. We propose OPDSearch+, the first
distillation paradigm that requires no teacher fine-tuning for search-augmented
reasoning. We investigate the role of a frozen off-the-shelf instruct model as the teacher in on-policy
distillation, and reveal a key insight: the teacher reshapes the student's policy distribution so that subsequent RL converges to a superior solution that RL alone cannot reach. In stage one, the student interacts with a live search engine and is distilled via a per-position forward KL objective, transferring
reasoning decomposition and evidence integration skills without any task-specific teacher training. In stage two, RL refines the distilled student from a richer behavioral foundation, achieving performance that RL alone cannot reach from scratch. Across seven QA
benchmarks, OPDSearch+ with a 3B model consistently outperforms all prior 3B RL baselines, achieving gains of 13.1% on HotpotQA and 8.5% on 2WikiMultihopQA.
Read the full abstract ▾
All authors Qinglin Ye, Zhiyuan Gu, Jingjie Xia, Yiheng Zhang, Kaiyan Zhao, Shunchao Zheng, Yuhang Mu, Wenchao Du, Yiming Wang
Artificial Intelligence · Cross-listed from cs.CL
SENSESHIFT: Continuous Sentiment-Controlled Text Generation via Encoder-based Mask Infilling
Shahed Masoudian, Markus Frohmann, Emmanouil Karystinaios et al. (+2 more)
Recent controllable text generation (CTG) for sentiment control has largely focused on decoder-based large
language models, making causal
attention the dominant paradigm. While effective for fluent generation, these models still struggle to satisfy complex constraints and follow fine-grained sentiment signals specified by users. Existing sentiment-aware CTG methods typically simplify the problem by treating sentiment either as a coarse categorical label (e.g., positive or negative) or as a single fine-grained control signal applied to an entire document. Consequently, more challenging settings such as sentence-level sentiment control within long-form text remain underexplored. To address these limitations, we introduce SenseShift , an encoder-based framework for fine-grained sentence-level CTG. Unlike standard decoder architectures, SenseShift leverages bidirectional
attention, quantized sentiment signals, and iterative mask infilling to generate local sentences conditioned on target sentiment intensity. Empirical evaluations on story and review generation demonstrate that SenseShift achieves stronger sentiment controllability while maintaining text quality and robustness to out-of-domain generation compared to larger decoder-based baselines.
Read the full abstract ▾
All authors Shahed Masoudian, Markus Frohmann, Emmanouil Karystinaios, Navid Rekabsaz, Markus Schedl
Artificial Intelligence
VideoHarness-RSI: Recursive Harness Self-Improvement for Long-Video Understanding with Frozen Vision-Language Models
Guoyang Xu and Hao Chen
Long-
video understanding depends not only on the capability of a
vision-
language model (VLM), but also on how its limited context is constructed from a much longer
video. Existing systems typically introduce hand-designed sampling,
retrieval,
memory, or agentic control strategies, making the context-construction program itself difficult to study as an independent optimization target. We introduce VideoHarness-RSI, a controlled framework that recursively searches executable context constructors around a frozen VLM while keeping the answering model and interface fixed. We study this baseline under complementary weak- and strong-initialization regimes. From a weak uniform constructor, recursive search progressively discovers more structured context-construction programs; from a stronger AKS harness, the same process further advances an already competitive hand-crafted frontier. The resulting harness retains its advantage under a matched cumulative visual-
token control and transfers directly to additional long-
video benchmarks without further search. Together, these results establish executable context construction as a distinct optimization layer and provide an auditable baseline for studying harness discovery, transfer, and efficiency around frozen VLMs.
Read the full abstract ▾
Artificial Intelligence · Cross-listed from cs.LG
Contrastive Branch Policy Optimization
Ying Wang, Changlin Qiu, Bang Lin et al. (+4 more)
Reinforcement learning with verifiable rewards (RLVR) enables
language models to learn multi-turn interaction with external tools, yet its sparse outcome rewards provide no signal for identifying which intermediate decisions are responsible for success. Branch sampling induces local comparisons among alternative continuations, but existing methods tend to conflate two distinct problems: allocating a fixed rollout budget and translating branch outcomes into
token-level credit. We introduce Contrastive Branch Policy Optimization (CBPO), which disentangles these two problems and assigns a dedicated mechanism to each. Generation
entropy screens candidate branch positions across the entire response, while path-level and node-level decay distribute a fixed budget across trajectories and positions to prevent exploration from collapsing onto a few paths or adjacent
tokens. A parent trajectory together with the branches that share an identical
token prefix forms an exact-prefix group, and the reward variation within this controlled group defines the Contrastive Branch Value (CBV), an outcome-based estimate of local decision sensitivity that rescales continuation advantages without altering their sign. When multiple nodes are selected along the same trajectory, CBPO partitions it into non-overlapping credit segments, thereby avoiding duplicated gradients on shared
tokens. Requiring only outcome rewards and no process-level annotation, CBPO provides a practical solution for fine-grained credit assignment in tool-integrated
agent training. Extensive experiments on ten
benchmarks, including five for mathematical
reasoning and five for knowledge-intensive search, show that CBPO consistently outperforms state-of-the-art policy-optimization and branch-based methods, attaining the highest macro-average accuracy in both domains and across two model scales.
Read the full abstract ▾
All authors Ying Wang, Changlin Qiu, Bang Lin, Linbo Jin, Wen Jiang, Zhe Sun, Jingli Yang